OER·harvester

Competency catalogue

The 104 active T2.3 competencies: the vocabulary the harvester searches with, the descriptor the alignment reads, and what the alignment found. Compose a harvest →

What the alignment foundHow often each was searched

How many resources the alignment reads as contributing to each competency or developing it: decided at level 2 or 3, or not decided yet and more likely than not to reach 2. 54 competencies have none, 24 one or two. Every figure is a lower bound: on a sample graded on every competency, the map counts 33% of what was graded 2 or 3, and almost nothing of the 5 competencies underlined with dots.

How many times a harvest has searched for each competency, once for each source it asked. 101 competencies have never been searched. A search is not material found: the aligned view says what came of it.

none 1–2 3–9 10–29 30+
never 1–2 3–5 6–11 12+
DDigital 94212203420012149086012303134 GGreen 0000000102000010000100010 RResilience & Entrepreneurial 004000000000000010000120000 TDeep Tech 72643202372825212335201140010

T21 Deep Tech · Skill · Intermediate

Bias Detection & Mitigation

Identify and address bias in data and model outputs

Alignment

1 resource likely to contribute to it or develop it

0 decided at level 2 or 3 · 1 not decided yet, more likely than not

On a graded sample, 4 pairs were graded 2 or 3 for it, and the map counts 0.

Retrieval

0 resources harvested by searching its terms

Never searched yet.

Search vocabulary

Domain group
bias detectionbias mitigationalgorithmic biasdataset biasfairness metricdebiasingdisparate impactmodel fairness

Zenodo and arXiv take the whole group in one query. GitHub takes one phrase at a time and works left to right, stopping once the competency is satisfied — so on that lane this order decides which terms actually run.

The compiled queries
zenodo ("artificial intelligence" OR "machine learning" OR "generative AI" OR "deep learning" OR "reinforcement learning" OR "large language model" OR "AI") AND ("bias detection" OR "bias mitigation" OR "algorithmic bias" OR "dataset bias" OR "fairness metric" OR "debiasing" OR "disparate impact" OR "model fairness")
arxiv (all:"artificial intelligence" OR all:"machine learning" OR all:"generative AI" OR all:"deep learning" OR all:"reinforcement learning" OR all:"large language model") AND (all:"bias detection" OR all:"bias mitigation" OR all:"algorithmic bias" OR all:"dataset bias" OR all:"fairness metric" OR all:"debiasing" OR all:"disparate impact" OR all:"model fairness")
github · term 1 AI "bias detection" in:name,description,readme fork:false archived:false
github · term 2 AI "bias mitigation" in:name,description,readme fork:false archived:false
github · term 3 AI "algorithmic bias" in:name,description,readme fork:false archived:false
github · term 4 AI "dataset bias" in:name,description,readme fork:false archived:false
github · term 5 AI "fairness metric" in:name,description,readme fork:false archived:false
github · term 6 AI "debiasing" in:name,description,readme fork:false archived:false
github · term 7 AI "disparate impact" in:name,description,readme fork:false archived:false
github · term 8 AI "model fairness" in:name,description,readme fork:false archived:false

Profile e461f35991a4cbb7

Alignment descriptor draft

Outcomes
  • Can detect bias in data and model outputs with fairness metrics.
  • Can apply mitigation techniques before, during or after training.
  • Can explain the trade-offs between fairness definitions.
Develops it

Hands-on material on bias detection and mitigation, where the learner measures bias in a dataset or model and reduces it.

Only mentions it

A warning that models 'may be biased'.

Alignment terms
fairnessbias mitigationfairness metricsdemographic parityequalized oddsfairlearnaif360algorithmic bias

Read in a resource's text beside the domain group. They are never compiled into a search.

False friends
bias termbias and variance

A term found beside one of these is read as another sense of the words, not as this competency.

Neighbours

Descriptor 3f40220eb1d5

How a vocabulary becomes a search

Every candidate must match both concept groups in its title, description or keywords before any content is downloaded. Precision warnings are recorded for audit; they never silently cancel positive evidence.

  1. Official name and description
  2. AI/ML terms and domain terms
  3. Zenodo, arXiv and GitHub syntax
  4. Metadata relevance gate
Mandatory AI/ML group
artificial intelligencemachine learninggenerative AIdeep learningreinforcement learninglarge language modelAI

GitHub's repository search takes one phrase at a time. Measured against the live endpoint, a parenthesised OR group returns nothing at all where each term alone returns thousands, and the 256-character limit rejects a long group outright — so that lane issues one query per term and stops as soon as the competency is satisfied.

coverage-vocabulary-1.3.0+sha256:08ad2cd19868 · catalogue sha256:04970bfc77255b1f

What a descriptor is

A descriptor is how the alignment reads a competency: what a learner who has it can do, what a resource has to do to develop it, what only mentions it, the phrases its subject is written in, its false friends, and how to tell it from the competencies that share its words. The definition above it stays the authority.

Every descriptor is a draft until a partner approves it; 0 of 104 are approved. Read the alignment rubric →

competency-descriptors-1.0.0+sha256:3cc741a61ced